What Drives Mortality Among HIV Patients in a Conflict Setting?

Crellen et al. Conflict and Health (2019) 13:52 https://doi.org/10.1186/s13031-019-0236-7

RESEARCH Open Access What drives mortality among HIV patients in a conflict setting? A prospective cohort study in the Central African Republic Thomas Crellen1,2* , Charles Ssonko3,TuridPiening4, Marcel Mbeko Simaleko5, Karen Geiger1 and M. Ruby Siddiqui3

Abstract Background: Provision of antiretroviral therapy (ART) in conflict settings is rarely attempted and little is known about the expected patterns of mortality. The Central African Republic (CAR) continues to have a low coverage of ART despite an estimated 110,000 people living with HIV and 5000 AIDS-related deaths in 2018. We present results from a cohort in Zemio, Haut-Mboumou prefecture. This region had the highest prevalence of HIV nationally (14.8% in a 2010 survey), and was subject to repeated attacks by armed groups on civilians during the observed period. Methods: Conflict from armed groups can impact cohort mortality rates i) directly if HIV patients are victims of armed conflict, or ii) indirectly if population displacement or fear of movement reduces access to ART. Using monthly counts of civilian deaths, injuries and abductions, we estimated the impact of the conflict on patient mortality. We also determined patient-level risk factors for mortality and how the risk of mortality varies with time spent in the cohort. Model-fitting was performed in a Bayesian framework, using logistic regression with terms accounting for temporal autocorrelation. Results: Patients were recruited and observed in the HIV treatment program from October 2011 to May 2017. Overall 1631 patients were enrolled and 1628 were included in the analysis giving 48,430 person-months at risk and 145 deaths. The crude survival rate after 12 months was 0.92 (95% CI 0.90, 0.93). Our model showed that patient mortality did not increase during periods of heightened conflict; the odds ratios (OR) 95% credible interval (CrI) for i) civilian fatalities and injuries, and ii) civilian abductions on patient mortality both spanned unity. The risk of mortality for individual patients was highest in the second month after entering the cohort, and declined seven- fold over the first 12 months. Male sex was associated with a higher mortality (odds ratio 1.70 [95% CrI 1.20, 2.33]) along with the severity of opportunistic infections (OIs) at baseline (OR 2.52; 95% CrI 2.01, 3.23 for stage 2 OIs compared with stage 1). Conclusions: Our results show that chronic conflict did not appear to adversely affect rates of mortality in this cohort, and that mortality was driven predominantly by patient-specific risk factors. The risk of mortality and recovery of CD4 T-cell counts observed in this conflict setting are comparable to those in stable resource poor settings, suggesting that conflict should not be a barrier in access to ART. Keywords: Antiretroviral therapy, Conflict setting, Survival analysis, Cohort study, Central African Republic

* Correspondence: [email protected] 1Médecins Sans Frontières Hollande, Avenue Barthelemy Boganda, PK4, Bangui, BP 1793, Central African Republic 2Mahidol-Oxford Tropical Medicine Research Unit, 420/6 Rajvithi Road, Tungphyathai, Bangkok 10400, Thailand Full list of author information is available at the end of the article

© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Crellen et al. Conflict and Health (2019) 13:52 Page 2 of 11

Introduction up (LTFU) of patients on ART in conflict settings after 12 Sub-Saharan Africa (SSA) exhibits disproportionate levels months was 9.0 and 8.1% respectively, which are within of morbidity and mortality from both infectious diseases the range of mortality and LTFU estimates from non- and conflict. Despite SSA accounting for 14.2% of the conflict settings [19]. world’spopulation[1], estimates by UNAIDS for 2018 The Central African Republic (CAR) is one of the gave the numbers of people living with HIV (PLHIV) in world’s least developed nations [20] and faces a number SSA at 25.6 million and AIDS-related deaths at 470,000; of challenges in healthcare provision including disrup- representing 67.5 and 61.0% respectively of the global to- tion of services due to instability and civil conflict along tals [2]. The Uppsala Conflict Data Program reported 20 with a lack of infrastructure, particularly outside of the separate conflicts in SSA in 2018; the African region had capital Bangui. At the onset of the study taking place, the highest number of fatalities from non-state conflicts CAR had an estimated 120,000 PLHIV and 11,000 (5408; 45.7%) and attacks against civilians (2888; 70.0%) AIDS-related deaths in 2013 [21] giving a mortality rate [3]. This is consistent with the disproportionate levels of of 91 deaths per 1000 PLHIV annually, which was the violent conflict in SSA since 2010 [4, 5]. Examples of re- highest in the world [22]. More recent estimates by gions in SSA with a dual burden of HIV and conflict in UNAIDS for 2018 suggest there are 110,000 PLHIV in 2018 include South Sudan, which had a national adult CAR (uncertainty interval [UI] 90,000 – 140,000) and prevalence of 2.5% in 2018 [2]thoughhigherestimates 4800 AIDS-related deaths (UI 3700 – 6400) [2]. In 2018, (3.1–6.8%) in conflict affected southern states [6], the it was estimated that 36% (UI 30–45%) of people living Cabo Delgado province of northern Mozambique which with HIV were receiving ART [2]. was subject to attacks by Islamic extremists in 2018 and MSF conducted a nationwide HIV seroprevalence study had a HIV prevalence of 12.6% in the same year [2], and in 2010 which found the Haut-Mbomou prefecture to Maniema province of the eastern Democratic Republic of have the highest prevalence in CAR; 14.8% compared with Congo, which saw clashes between government forces and a prevalence of 5.9% nationally among individuals 15–49 rebel groups in 2018, had an estimated HIV prevalence of years of age [23]. Access to treatment in CAR was particu- 3.4% compared with a national average of 1.0% [7]. larly low at this time, 13.8% in 2010 [21]. The health Nations in SSA affected by conflict have been shown to centre in Zemio, Haut-Mboumou prefecture has been have worse health outcomes, such as rates of mortality for supported by MSF since 2010 and an ART cohort was mothers and children under-five [8, 9]. The relationship be- established in October 2011. This was initially a response tween conflict and the transmission of infectious disease is to an influx of Congolese refugees and internally displaced more complex and contested in the literature. Conflict in persons (IDPs), who had fled their homes due to repeated SSA between 1997 and 2010 did not appear to result in attacks by the Lord’s Resistance Army (LRA) during higher transmission rates for falciparum malaria and the 2008–10. The region has since faced repeated violent in- pre-conflict trend was generally maintained [10]. For HIV, cursions from armed groups [24]. The Zemio HIV treat- a number of multi-site studies in SSA have argued there is ment program therefore represented an attempt to no evidence that infection incidence increases during con- provide access to ART for a population with a high disease flict when compared with pre-conflict rates [11–13]. How- burden in a conflict setting. ever, a cohort study of police officers in Guinea-Bissau Reports of HIV patient outcomes generally focus on during the 1998–9 civil war reached the opposite conclu- summary statistics after defined intervals (6 or 12 sion for the incidence of HIV-1 [14], and the observations months) and are rarely explored as a detailed time series. of stable or decreasing HIV incidence before and during It is important to consider temporal variation in patient conflict from cross-sectional surveys could be confounded mortality as this has implications for management of pa- by higher rates of, unobserved, AIDS-related mortality [15]. tient care and allocation of resources during ART pro- Provision of anti-retroviral therapy (ART) during con- grams. For instance, the risk of mortality among patients flict has, historically, not been prioritised due to a percep- in sub-Saharan Africa is particularly high in the first 3 tion that this was too difficult to achieve [16]. In the face months of ART provision [25, 26] due to individuals of long term conflicts in areas with a high HIV burden, seeking treatment at a later stage of disease progression there was concern in the humanitarian community that [27]. In the case of studies on the impact of conflict, lon- “treatment cannot wait” [13, 17]. In 2010 the medical or- gitudinal data is particularly important as instability is ganisation Médecins Sans Frontières (MSF) reported re- rarely temporally uniform and often consists of periods sults from 22 ART programs in conflict or post-conflict of relative calm punctuated by acute violence [17]. Con- settings in SSA, finding that patient outcomes were com- flict from armed groups can impact mortality rates i) parable to those in stable resource-limited settings [18]. directly if HIV patients are victims of armed conflict, or These findings were also shown in a systematic review ii) indirectly if population displacement or fear of move- and meta-analysis that found mortality and loss to follow- ment reduces adherence to ART [28]. Crellen et al. Conflict and Health (2019) 13:52 Page 3 of 11

In this study, we address both temporal variation in 2011 to 2012 for patients with CD4 < 350 cells/μL, from survival rates and the impact of conflict on patient mor- 2013 to 2016 with < 500 cells/μL and for all patients regard- tality. We use a logistic regression survival analysis less of CD4 count from 2017 (universal test and treat). First where the underlying rate of mortality is permitted to line therapy was typically tenofovir/lamivudine/ efavirenz vary by month. The impact of the conflict is measured (TDF/3TC/EFV) combination therapy, second line therapy though aggregated monthly counts of civilian fatalities, was zidovudine/lamivudine/ lopinavir/ritonavir (AZT/3TC/ injuries and abductions from armed groups. We deter- LPVr). Patients aged ≤15 years old were treated with paedi- mine associations between the intensity of conflict with atric drug formulations. Patients with CD4 counts above the risk of monthly mortality, in addition to exploring the threshold were given prophylactic treatment against op- the effect of patient-level risk factors. portunistic infections (OIs) and retested every 6 months untiltheywereeligibletoinitiate.Co-trimoxazoleprophy- Methods laxis was given to all patients before the initiation of ART This research fulfilled the exemption criteria set by the and continued during ART. Patients were given appoint- MSF Ethical Review Board (ERB) for a posteriori ana- mentsatleastevery3monthswhenARTwasprovidedand lyses of routinely collected clinical data and thus did not clinical signs of OIs were recorded and treated. All diagnos- require MSF ERB review. This study was conducted with tics, treatment and appointments were provided to patients permission from the Medical Director Sidney Wong free of charge by MSF. (MSF-Operational Centre Amsterdam). Patients in the The Zemio HIV program was mainly nurse driven, an Zemio HIV cohort gave written consent for anonymised expatriate nurse was present to provide training and guid- data to be used for research purposes. ance to two national staff nurse consultants and secour- We used data from the treatment program’s commence- istes (trained lay workers) who provided HIV testing and ment on 18th October 2011 until 31st May 2017. Patients counseling and psychosocial care to patients. HIV test re- were recruited prospectively throughout the study period sults were confirmed by a lab technician supported by and were drawn from the surrounding area, which in- secouristes. A medical doctor was consulted from time to cludes the Bas-Uele province of neighbouring Democratic time to provided support on patients with complications. Republic of Congo along with Congolese refugee camps The Zemio team was also supported by the HIV/TB ad- and IDP camps around Zemio (Fig. 1). The target popula- viser, health adviser and epidemiologist from MSF head- tion consisted mainly of the local Azande people who quarters. For the supply of ART, MSF ordered ARVs and work predominantly as subsistence farmers (Table 1), and OI drugs internationally and delivered them to Zemio by Fulani pastoralists who also reside in the region. Patients plane until 2015, after which supplies were partly ordered qualified for inclusion in the study i) if they volunteered through Global Fund in Bangui. An important aspect of for HIV testing and counselling at the Zemio health centre the program was community engagement, MSF used 50 (usually following contact with community health community health workers to build acceptance of the pro- workers), ii) if they were referred to the testing and coun- gram, report on patient outcomes and encourage patients selling program after an appointment in Zemio or nearby to remain on ART. health facilities, or iii) during the first antenatal visit for Mortality during the study was reported directly if pa- pregnant women. Infants testing positive for HIV after tients died in hospital or if the patient died in the com- birth were also enrolled into the study (Fig. 2). The study munity this was reported by community health workers used data from medical records collected during patient who were in frequent contact with patients. As the cause enrolment and consultations, which were entered into a of death was not always clearly reported or may have modified Microsoft Access database (FUCHIA) used by been unknown, we have considered all deaths to be MSF for monitoring of HIV cohorts. HIV/AIDS related. Loss to follow-up was defined as liv- Patients were tested for HIV with the Determine™ HIV- ing patients ≥6 months absent from their last appoint- 1/2 (Alere) antigen/antibody test and Uni-Gold™ Recombi- ment at the end of the study period. gen® (Trinity Biotech) antibody test using whole blood; a We used a logistic regression model for the survival positive diagnosis from both tests was required before diag- analysis, where the outcomes and time-varying covari- nosing a patient as positive and the tests were repeated in ates were aggregated by month. The probability of mor- the case of discordant results. Infants were tested with PCR tality for patient j in month i was given by a logit- every 6 months up to 18 months after birth either i) from a transformed linear function of an intercept and covari- dry blood spot sent to Global Clinical and Viral Laboratory, ates. Covariates were i) sex, ii) patient’s age at cohort South Africa prior to 2017 or ii) using Xpert® HIV-1 Qual entry in years, iii) clinical stage of OIs, iv) the number of (Cepheid) from whole blood on-site in 2017. civilian fatalities and injuries per month from armed The Zemio program offered anti-retrovirals (ARVs) to groups, and v) abductions per month from armed individuals based on WHO guidelines [29–31]; from groups in the patient catchment area, as recorded by the Crellen et al. Conflict and Health (2019) 13:52 Page 4 of 11

Fig. 1 (See legend on next page.) Crellen et al. Conflict and Health (2019) 13:52 Page 5 of 11

(See figure on previous page.) Fig. 1 The Central African Republic (CAR) is shown in a regional context (a), with the country divided by prefectures. The study area is indicated by the box in the south east of CAR. The town of Zemio is shown in context (b), within Haut-Mbomou prefecture and adjacent to Mbomou prefecture. Nearby towns where patients were drawn from for the HIV treatment program (Rafai, Djema and Mboki) and the Bas-Uele province of the Democratic Republic of the Congo are also shown. The majority of patients were drawn from around the town of Zemio, including internally displaced person and refugee camps resulting from conflict in 2010 by the Lord’s Resistance Army. Grey lines indicate the road network in the CAR. The number of patients recruited into the cohort from each location is indicated by the size of the points (n = 1631). A time series of civilian deaths / injuries and abductions by armed groups in the patient catchment area details the evolution of the conflict in this region from October 2011 – May 2017 (c)

NGO Invisible Children [24]. The model intercept was distributions as the measure of central tendency and the permitted to vary by the number of months since the pa- 95% credible intervals (CrI) as the measure of dispersion. tient entered the cohort, where the monthly intercepts were drawn from a multivariate normal distribution with a covariance matrix which accounted for temporal auto- Results correlation between months [32]. A total of 1631 HIV positive patients were recruited into Data cleaning and descriptive analysis was performed the study, 1147 were female (70.3%) and the median age using R (version 3.4.4) and the models were fitted with at first visit was 29 years (interquartile range [IQR] 22, Hamiltonian Markov Chain Monte Carlo (MCMC) using 37). All patients were initiated on co-trimoxazole and Stan (v2.17.3). We assigned vaguely informative normal the majority were enrolled on ART (n = 1491, 91.4%). In prior distributions to parameters [33]. Four parallel total there were 148 deaths (9.1%), of which the major- MCMC chains were run for 40,000 iterations including ity, 121 (81.8%), were among patients on ART. Patients burn-in, and convergence was assessed using the Gelman- spent a median of 27 months in the cohort (IQR 12, 48). Rubin statistic, the effective sample size and visual inspec- The proportion of patients who survived the first 12 tion [34]. We used the median of posterior parameter months in the cohort was 1212/1315; 0.92 (95% confidence interval [CI] 0.90, 0.93). Overall 183/1631 (11.2%) Table 1 Characteristics of the 1631 HIV-positive patients patients met the criteria for loss to follow up. Patients n enrolled in the Zemio HIV cohort, Central African Republic from that were CD4 tested on their first visit ( = 1017) had a μ October 2011–May 2017. Interquartile range = IQR, ART = median count of 268 CD4 cells/ l (IQR 148, 427), this antiretroviral therapy, OIs = opportunistic infections. Loss to rose after 6 months on ART to a median of 403 CD4 follow up is defined as patients not present for an appointment cells/μl (IQR 240, 634; n = 442) and after 12 months to a > 12 months before the end of the program median of 433 CD4 cells/μl (IQR 264, 636; n = 321). Pa- Total patients enrolled in the Zemio 1631 (4107 person-years tient characteristics are summarised in Table 1. HIV cohort at risk) For the statistical analysis, we excluded three patients Patients enrolled on ART 1491/1631 (91.4%) who had incomplete data (all three were missing both Female patients 1147/1631 (70.3%) the date of enrolment and clinical staging of OIs at base- Median age at first appointment (IQR) 29 years (22, 37) line). This therefore left 1628 patients who contributed 48,430 person-months at risk (Fig. 2). Modelling patient Deaths 148 mortality as a series of monthly Bernoulli trials, the Loss to follow up 183 baseline monthly risk of mortality declined the longer CD4 cell count/μl, baseline (IQR), n = 1017 268 (148, 427) the patient remained in the cohort. In the first month a CD4 cell count/μl, 6 months (IQR), n = 442 403 (240, 634) patient entered the cohort, the median baseline risk CD4 cell count/μl, 12 months (IQR), n = 321 433 (264, 636) (probability when all covariates are set to zero) of mor- − 4 − 4 − 3 Clinical Stage 1 OIs at baseline 241/1628 (14.8%) tality was 9.5 × 10 (95% CrI 4.1 × 10 , 2.1 × 10 ). The risk of mortality peaked in the second month at Clinical Stage 2 OIs at baseline 403/1628 (24.7%) − 0.0010 (95% CrI 4.7 × 10 4, 0.0023) and subsequently Clinical Stage 3 OIs at baseline 789/1628 (48.5%) declined nearly seven-fold by the twelfth month to 1.5 × − − − Clinical Stage 4 OIs at baseline 195/1628 (12.0%) 10 4 (95% CrI 5.8 × 10 5, 3.9 × 10 4). We show the risk Occupation: Infants/ Pupil / Student 148/1628 (9.1%) of mortality by month in the cohort in Fig. 3, where the Occupation: Farmers / Fishermen / Domestic 1272/1628 (78.1%) risk has been scaled by the covariates for an “average” Occupation: Small commerce 139/1628 (8.5%) patient (female, 31 years old, OI stage 3 at baseline, 2 deaths/injuries per month, 21 abductions per month); of Occupation: Civil servant/ official 26/1628 (1.6%) note is the rapid decline in the monthly risk of mortality Occupation: Other /not recorded 43/1628 (2.6%) over the first year the patient is in the cohort. Crellen et al. Conflict and Health (2019) 13:52 Page 6 of 11

Fig. 2 Flow chart for patient recruitment into the Zemio HIV cohort, Central African Republic between October 2011 – May 2017 and inclusion into the statistical analysis. The figures for patients started on ART / not started on ART and deaths are from those included in the statistical analysis

The impact of covariates on the baseline risk of mor- odds of mortality was the severity of opportunistic infec- tality is shown in Table 2 as odds ratios. Male sex was tions (OIs) at baseline. Relative to individuals with stage associated with a higher monthly risk of mortality; me- 1 OIs, stage 2 OIs had an OR of 2.5 (95% CrI 2.0, 3.2), dian odds ratio (OR) 1.7 (95% CrI 1.2, 2.3). Higher age stage 3 OIs had an OR of 6.4 (95% CrI 4.1, 10) and stage was also associated with an increased odds of mortality; 4 OIs had an OR of 16 (95% CrI 8.2, 34). The time- OR 1.01 (95% CrI 1.00, 1.03) for each unit (year) in- varying conflict covariates had a weaker association with crease. The covariate with the highest impact on the monthly patient mortality, with the credible interval

Fig. 3 The monthly risk of mortality for an average HIV patient in the Zemio cohort. The point gives the posterior median, the broad bar the 80% posterior credible interval and the thin bar the 95% posterior credible interval. An average patient is defined as a female, 31 years old at baseline, with clinical stage 3 opportunistic infections at baseline. We also assume 2 civilian deaths or injuries per month and 21 abductions (median values for conflict explanatory variables). The risk varies by the length of time the patient is in the cohort, with the probability of mortality highest in the first 2 months, and then declines rapidly until the twelfth month. Note that these monthly probabilities are conditional on the patient having survived until that point Crellen et al. Conflict and Health (2019) 13:52 Page 7 of 11

Table 2 Odds ratios of covariates on the risk of monthly mortality estimated from 48,430 patient months at risk from 1628 HIV- positive patients enrolled in the Zemio HIV cohort, Central African Republic from October 2011–May 2017 Covariate (range of values) Classification Median odds ratio 95% Credible Interval Patient Sex (male or female; female is baseline) Binary 1.70 1.20, 2.33 Age at entry (0–71 years) Continuous integer 1.01 1.00, 1.03 Clinical Staging of OIs (stage 1–4) Continuous integer 2.52 2.01, 3.23 Civilian deaths and injuries per month (0–153) Continuous integer 1.00 0.99, 1.01 Civilian abductions per month (0–119) Continuous integer 1.00 0.99, 1.00 spanning unity in both instances. Fatalities and injuries A meta-review of mortality during ART from 18 cohort from conflict had an OR of 1.00 (95% CrI 0.99, 1.01), studies in West, East and Southern Africa reported sur- and abductions had an OR of 1.00 (95% CrI 0.99, 1.00) vival at 12 months ranging from 0.74 to 0.92 [25], there- on the monthly risk of mortality per unit increase. fore our crude 12 month survival estimate of 0.92 (95% CI The cumulative risk of mortality over the length of 0.90, 0.93) is at the higher level of this range, and compar- time spent in the cohort is shown in Fig. 4, stratified by able to estimates from conflict settings such as Bukavu, A) patient sex and B) clinical staging (severity) of OIs at DRC, which was also estimated at 0.92 (95% CI 0.88, 0.96) baseline. The proportion of females (31 years old, OI [17]. This high survival rate may be partly explained by stage 3 at baseline, 2 deaths/injuries per month 21 ab- the median CD4 count we observe at baseline; 268 (IQR ductions per months) surviving after 60 months is esti- 148, 427; Table 1), which is higher than in other published mated to be 0.89 (95% CrI 0.86, 0.91) and for males with African studies and is comparable to baseline CD4 counts the same characteristics is 0.82 (95% CrI 0.77, 0.86). The observed in cohorts from developed countries (median of proportion of patients (females, 31 years old, 2 deaths/ 234 cells/μL from European and North American cohorts) injuries per month 21 abductions per months) surviving [25, 35]. Counts of CD4 cells are highly predictive of early after 60 months by OI severity at baseline are 0.98 (95% patient mortality; patients with < 50 cells/μLwerefound CrI 0.97, 0.99), 0.95 (95% CrI 0.94, 0.97), 0.89 (95% CrI to have a risk of mortality 2.5 times that of patients with 0.86, 0.91) and 0.74 (95% CrI 0.67, 0.81) for clinical CD4 counts ≥50 cells/μL after 12 months [25]. In our ana- stages 1, 2, 3 and 4 respectively. lysis we opted to use clinical staging as an explanatory The logistic regression model showed adequate con- variable for survival in lieu of CD4 counts as we were un- vergence according to best practice guidelines [34], with able to attain results for every patient owing to ruptures in all parameters showing an effective sample size > 400 stock and the difficulty of maintaining laboratory equip- and the Gelman-Rubin statistic ≤1.01, indicating that ment in a remote rural setting. The provision of co- chains had mixed well and run for sufficient iterations. trimoxazole alongside ART throughout the treatment period likely also contributed to high survival rates due to Discussion partial protection against OIs and malaria, the latter is the This study analysed the pattern of mortality among a co- leading cause of death in CAR [39, 40]. Mortality was hort of 1628 HIV patients in Zemio, Central African Re- highest in the first 3 months of enrolment and peaked in public over 48,430 person-months at risk and determined if the second month (Fig. 3), this is consistent with other co- conflict from armed groups in the surrounding area was as- hort studies in SSA that showed the highest rates of mor- sociated with a higher risk of patient mortality. The results tality in the first 3 months of treatment [25, 26]. from the logistic regression are strongly suggestive of no ef- The time series of civilian fatalities, injuries and abduc- fect from i) fatalities and injuries from armed groups, or ii) tions by numerous armed groups shows the evolution of a abductions from armed groups on civilians on the risk of long term conflict (Fig. 1c). There was a high underlying HIV patient mortality. Patient mortality was more strongly rate of civilian abductions throughout the period, predom- associated with both patient sex (OR 1.7 for males com- inantly by the LRA. Civilian mortality before 2017 was due pared with females) and clinical staging of OIs at baseline; mainly to attacks by the LRA or unidentified armed groups patients with OIs stage 4 had an OR of 16 compared with on civilians. During 2017 the region became embroiled in patients with stage 1 OIs, which is consistent with findings a larger national conflict between Christian Anti-Balaka from cohort studies in low income settings [35]. The higher and Muslim Ex-Seleka militias who clashed in the patient risk of mortality for men could be explained by poorer catchment area of Mboumou / Haut-Mboumou. Road ART retention rates or differences in health seeking behav- blocks by the LRA, in addition to a lack of public transport iour following diagnosis, which have been observed in other made it challenging for patients to travel to Zemio for ap- African settings [36–38], however we did not collect data pointments or ART collection. Armed robberies of the on these confounding variables in this study. staff compound were also a frequent occurrence and Crellen et al. Conflict and Health (2019) 13:52 Page 8 of 11

Fig. 4 The cumulative probability of survival for HIV patients in the Zemio cohort shown a by sex, and b by clinical stage of opportunistic infections at baseline. The thick line shows the posterior median and the dotted lines show the 95% posterior credible interval. The patients stratified by gender are given covariate values of 31 years old at baseline, clinical stage 3 opportunistic infections at baseline and we assume 2 civilian deaths or injuries per month and 21 abductions for each month. The patients stratified by clinical stage of OIs, are female, 31 years old at baseline, and we assume 2 civilian deaths or injuries per month and 21 abductions for each month fighting inside Zemio led to expatriate MSF staff being Furthermore, the relatively low numbers lost to follow-up evacuated twice over the study period [41]. Despite this in- (11.2% over 5 years) and the recovery of CD4-cell counts stability, the pattern of mortality among HIV patients is suggests that the Zemio HIV program was successful in similar to that observed in longitudinal studies in non- retaining patients on ART. conflict settings, where mortality rates are largely driven As in any cohort analysis with right censored observa- by the clinical severity of OIs at baseline [42]. tions, our results are subject to limitations. When Crellen et al. Conflict and Health (2019) 13:52 Page 9 of 11

unreported deaths are classified as loss to follow-up this of care was continued from December 2017 onwards may cause the mortality rate to be biased downwards. and MSF continues to monitor the impact of conflict on Conversely, the mortality rate may be inflated by consid- mortality and patient outcomes for HIV patients in ering all deaths that occurred in the cohort to be HIV/ Zemio. AIDs related. This was a conservative decision to com- pensate for the lack of information on causes of death, Conclusion particularly when patients died in the community ra- We have shown that the pattern of mortality among a ther than at the health centre. We considered the ef- cohort of HIV patients in the Central African Republic fect of conflict in the patient catchment area, is driven more strongly by patient level covariates than measured by civilian deaths/ injuries and abductions, by the evolution of the surrounding long term conflict. to be constant for all communities. Patients drawn The levels of mortality we observed in this conflict set- from Obo to the east of Zemio and those drawn from ting are comparable to those in stable resource poor set- the west from Rafaï are separated by 350 km, there- tings. We call for a renewed emphasis on increased fore a more nuanced approach would be to consider access to ART in conflict settings. the impact of conflict at smaller spatial scales. The Acknowledgements largest upsurge in violence against civilians also coin- Funding was provided by Médecins Sans Frontières. The authors would like cided with the end of the MSF program and the tran- to thank the members of the Zemio HIV team for their hard work, sition to a community led HIV program in Zemio professionalism and dedication to patients throughout the program: Mr. Peeters Muntulongo, Mr. Francois d’Assis Koyassama (Nurses), Mr. Romuald from mid-2017 (Fig. 1c); we were therefore unable to Gbiakota (Auxiliary Nurse), Mr. Joseph Mbolingbagbe, Ms. Francisca Zounga- assess the longer term impact of this interruption to Awo and Mr. Martin Gbaboundou (Counsellors), Mr. Alain Mbolihinipai, Mr. treatment. Nevertheless, our study is among the first Jacques Koumbodigui (Pharmacists), Mr. Hassan Mahamat (Data Encoder) and Mr. Igor Mokopa (Laboratory Technician). MSF operates with the support to quantitatively assess temporally varying conflict on of national and local HIV authorities, we therefore acknowledge the role of HIV cohort mortality and the large sample size gives the CAR Ministry of Health and Dr. Diemer Henri St. Calvaire in particular. We confidence to the parameter estimates we have would also like to thank Ms. Nastasia Stipo and Mr. Paul Ronan at Invisible Children for sharing their data on conflict in the Zemio region. In addition, obtained. we thank Prof. Ben S. Cooper, Dr. Mo Yin, Ms. Anastasia Hernandez Koutou- The generalisability of our findings is influenced by cheva, and Dr. Cono Ariti for their valuable comments on the manuscript. the representativeness of the cohort composition. Not- Authors’ contributions ably women far outnumber men in the Zemio cohort TC conceived and performed the analysis and wrote the manuscript. CS, TP (70.3%). Whether HIV prevalence is higher among and MRS conceived, monitored and were responsible for the treatment women in the community is unknown and depends on program, along with editing the manuscript. MMS and KG contributed to the treatment program and edited the manuscript. All authors have seen and the pattern of transmission [43, 44]. The majority of re- approved the final version of the manuscript. corded occupations among adults is agricultural workers, which is representative of the general popula- Funding tion; The World Bank estimates that around 85% of The study was funded through MSF core funding. adults in the CAR work as subsistence agriculturalists Availability of data and materials [45]. Health workers and medical staff in Zemio com- The data on aggregated monthly counts of patient mortality used in the first mented on the high level of acceptance of the HIV model and all code are available from the corresponding authors on request. Individual level patient data cannot be shared to protect patient treatment program by the community, which was likely confidentiality. reinforced by the community health workers and the free healthcare offered by MSF at the health centre in a Ethics approval and consent to participate Médecins Sans Frontières (MSF) is able to operate in countries such as the number of departments. In conflict settings where com- Central African Republic only with the support of national and local munities are less receptive to ART programs, the up- authorities and through continued dialogue with the beneficiary take and adherence to treatment, and subsequently the communities. Patients in the Zemio HIV cohort gave written consent for anonymised data to be used for research purposes. This research fulfilled the risk of mortality, may be heightened. exemption criteria set by the MSF Ethical Review Board (ERB) for a posteriori The Zemio HIV cohort transitioned to a community- analyses of routinely collected clinical data and thus did not require MSF ERB based model of care in 2017 in order to reduce operating review. This study was conducted with permission from the Medical Director Sidney Wong (MSF-Operational Centre Amsterdam). costs, personnel requirements, and ensure the future sustainability of the program. Community models have Consent for publication been shown in low-resource settings to deliver outcomes Not applicable. for patients that are comparable to provider-based care Competing interests [46, 47]. This transition was interrupted when fighting The authors declare that they have no competing interests. broke out again in the town of Zemio in July 2017 (at Author details the end of the observed period) leading MSF to tempor- 1Médecins Sans Frontières Hollande, Avenue Barthelemy Boganda, PK4, arily suspend its operations [48]. The community model Bangui, BP 1793, Central African Republic. 2Mahidol-Oxford Tropical Medicine Crellen et al. Conflict and Health (2019) 13:52 Page 10 of 11

Research Unit, 420/6 Rajvithi Road, Tungphyathai, Bangkok 10400, Thailand. 22. Granich R, Gupta S, Hersh B, Williams B, Montaner J, Young B, et al. Trends in 3Médecins Sans Frontières, The Manson Unit, Chancery Exchange, 10 Furnival AIDS deaths, new infections and ART coverage in the top 30 countries with Street, London EC4A 1AB, UK. 4Médecins Sans Frontières, Am Köllnischen the highest AIDS mortality burden; 1990–2013. PLoS One. 2015;10(7):e0131353. Park 1, 10179 Berlin, Germany. 5Ministre de la Santé Publique et de la 23. Médecins Sans Frontières. Central African Republic: a state of silent crisis. Population, Bangui, BP 883, Central African Republic. Amsterdam: MSF; 2011. 24. Invisible Children. LRA data tracker 2017. Available from: https://www. Received: 19 March 2019 Accepted: 18 October 2019 lracrisistracker.com/. 1 July 2017. 25. Lawn SD, Harries AD, Anglaret X, Myer L, Wood R. Early mortality among adults accessing antiretroviral treatment programmes in sub-Saharan Africa. AIDS (London, England). 2008;22(15):1897–908. References 26. Harries AD, Schouten EJ, Libamba E. Scaling up antiretroviral treatment 1. World Bank. Population data. Available from: data.worldbank.org. Accessed 5 in resource-poor settings. Lancet. 2006;367(9525):1870–2. Aug 2019. 27. Zachariah R, Fitzgerald M, Massaquoi M, Pasulani O, Arnould L, Makombe S, 2. UNAIDS. Global HIV & AIDS statistics. Available from: https://www.unaids.org. et al. Risk factors for high early mortality in patients on antiretroviral Accessed 5 Aug 2019. treatment in a rural district of Malawi. AIDS. 2006;20(18):2355–60. 3. Uppsala Conflict Data Program. Data export 2018. Available from: https:// 28. Pyne-Mercier LD, John-Stewart GC, Richardson BA, Kagondu NL, Thiga J, ucdp.uu.se/. Accessed 5 Aug 2019. Noshy H, et al. The consequences of post-election violence on 4. Themnér L, Wallensteen P. Armed conflicts, 1946–2011. J Peace Res. 2012; antiretroviral HIV therapy in Kenya. AIDS Care. 2011;23(5):562–8. 49(4):565–75. 29. World Health Organization. Guideline on when to start antiretroviral therapy and 5. Williams PD. Continuity and change in war and conflict in Africa. Prism. on pre-exposure prophylaxis for HIV. Geneva: World Health Organization; 2015. 2017;6(4):32–45. 30. Volberding PA, Deeks SG. Antiretroviral therapy and management of HIV 6. Jervase A, Tahir H, Modi JK, Almobarak AO, Mital D, Ahmed MH. HIV/AIDS in infection. Lancet. 2010;376(9734):49–62. South Sudan past, present, and future: a model of resilience in a 31. World Health Organization. Consolidated guidelines on the use of antiretroviral challenging context. J Public Health Emerg. 2018;2:31. drugs for treating and preventing HIV infection: recommendations for a public 7. Kokolomami JHT, Kayembe PK. HIV / AIDS epidemic in the Democratic health approach. Geneva: World Health Organization; 2016. Republic of the Congo: current level of key indicators and projection by 32. McElreath R. Statistical rethinking: a Bayesian course with examples in R and 2030. Cent Afr J Public Health. 2018;4(3):86–94. Stan. Boca Raton: Chapman and Hall/CRC; 2018. 8. O'hare BA, Southall DP. First do no harm: the impact of recent armed 33. Gelman A, Carlin JB, Stern HS, Rubin DB. Bayesian data analysis. Boca Raton: conflict on maternal and child health in sub-Saharan Africa. J R Soc Med. Taylor & Francis; 2014. 2007;100(12):564–70. 34. Vehtari A, Gelman A, Simpson D, Carpenter B, Bürkner P-C. Rank- 9. Boerma T, Tappis H, Saad-Haddad G, Das J, Melesse DY, DeJong J, et al. normalization, folding, and localization: An improved R-hat for assessing Armed conflicts and national trends in reproductive, maternal, newborn convergence of MCMC. arXiv preprint arXiv:190308008; 2019. and child health in sub-Saharan Africa: what can national health surveys tell 35. Braitstein P, Brinkhof MW, Dabis F, Schechter M. Mortality of HIV-1-infected us? BMJ Glob Health. 2019;4(Suppl 4):e001300. patients in the first year of antiretroviral therapy: comparison between low- 10. Sedda L, Qi Q, Tatem AJ. A geostatistical analysis of the association income and high-income countries. Lancet. 2006;367(9513):817. between armed conflicts and plasmodium falciparum malaria in Africa, 36. Braitstein P, Boulle A, Nash D, Brinkhof MW, Dabis F, Laurent C, et al. Gender 1997–2010. Malar J. 2015;14(1):500. and the use of antiretroviral treatment in resource-constrained settings: 11. Spiegel PB, Bennedsen AR, Claass J, Bruns L, Patterson N, Yiweza D, et al. findings from a multicenter collaboration. J Women's Health. 2008;17(1):47–55. Prevalence of HIV infection in conflict-affected and displaced people in seven sub- 37. Brinkhof MW, Boulle A, Weigel R, Messou E, Mathers C, Orrell C, et al. Mortality Saharan African countries: a systematic review. Lancet. 2007;369(9580):2187–95. of HIV-infected patients starting antiretroviral therapy in sub-Saharan Africa: 12. Bennett BW, Marshall BD, Gjelsvik A, McGarvey ST, Lurie MN. HIV incidence comparison with HIV-unrelated mortality. PLoS Med. 2009;6(4):e1000066. prior to, during, and after violent conflict in 36 sub-Saharan African nations, 38. Druyts E, Dybul M, Kanters S, Nachega J, Birungi J, Ford N, et al. Male 1990-2012: an ecological study. PLoS One. 2015;10(11):e0142343. sex and the risk of mortality among individuals enrolled in antiretroviral 13. Spiegel PB. HIV/AIDS among conflict-affected and displaced populations: therapy programs in Africa: a systematic review and meta-analysis. AIDS. dispelling myths and taking action. Disasters. 2004;28(3):322–39. 2013;27(3):417–25. 14. Månsson F, Biague A, da Silva ZJ, Dias F, Nilsson LF, Andersson S, et al. 39. Walker A, Ford D, Gilks C, Munderi P, Ssali F, Reid A, et al. Daily co- Prevalence and incidence of HIV-1 and HIV-2 before, during and after a civil trimoxazole prophylaxis in severely immunosuppressed HIV-infected war in an occupational cohort in Guinea-Bissau, West Africa. AIDS. 2009; adults in Africa started on combination antiretroviral therapy: an 23(12):1575–82. observational analysis of the DART cohort. Lancet. 2010;375(9722): 15. Shannon K, Kaida A, Rachlis B, Lloyd-Smith E, Gray G, Strathdee SA. 1278–86. Reconsidering the impact of conflict on HIV infection among women in the 40. Sandison TG, Homsy J, Arinaitwe E, Wanzira H, Kakuru A, Bigira V, et al. Protective era of antiretroviral treatment scale-up in sub-Saharan Africa: a gender lens. efficacy of co-trimoxazole prophylaxis against malaria in HIV exposed children in AIDS. 2008;22(14):1705–7. rural Uganda: a randomised clinical trial. BMJ. 2011;342:d1617. 16. Ellman T, Culbert H, Torres-Feced V. Treatment of AIDS in conflict-affected 41. Médecins Sans Frontières. MSF suspends humanitarian relief activities settings: a failure of imagination. Lancet. 2005;365(9456):278–80. following attack in Bangassou 2017. Available from: https://www.msf.org/ 17. Culbert H, Tu D, O'Brien DP, Ellman T, Mills C, Ford N, et al. HIV treatment in central-african-republic-msf-suspends-humanitarian-relief-activities-following- a conflict setting: outcomes and experiences from Bukavu, Democratic attack-bangassou. 16 Sept 2018. Republic of the Congo. PLoS Med. 2007;4(5):e129. 42. Lawn SD, Little F, Bekker L-G, Kaplan R, Campbel E, Orrell C, et al. 18. O'Brien DP, Venis S, Greig J, Shanks L, Ellman T, Sabapathy K, et al. Changing mortality risk associated with CD4 cell response to Provision of antiretroviral treatment in conflict settings: the experience antiretroviral therapy in South Africa. AIDS (London, England). 2009; of Médecins Sans Frontières. Confl Health. 2010;4(1):12. 23(3):335. 19. Griffiths K, Ford N. Provision of antiretroviral care to displaced populations 43. De Oliveira T, Kharsany AB, Gräf T, Cawood C, Khanyile D, Grobler A, et al. in humanitarian settings: a systematic review. Med Conflict Survival. 2013; Transmission networks and risk of HIV infection in KwaZulu-Natal, South 29(3):198–215. Africa: a community-wide phylogenetic study. Lancet HIV. 2017;4(1):e41–50. 20. United Nations Development Program. 2016 human development 44. Dellar RC, Dlamini S, Karim QA. Adolescent girls and young women: key report 2017. Available from: http://hdr.undp.org/en/2016-report.9Aug populations for HIV epidemic control. J Int AIDS Soc. 2015;18:19408. 2017. 45. World Bank. DataBank reports 2019. Available from: https://databank. 21. Joint United Nations Programme on HIV/AIDS. Hundreds of people living worldbank.org/. 30 Aug 2019. with HIV resume treatment in the Central African Republic 2015. Available 46. Bemelmans M, Baert S, Goemaere E, Wilkinson L, Vandendyck M, Cutsem G, from: http://www.unaids.org/en/resources/presscentre/featurestories/2015/ et al. Community-supported models of care for people on HIV treatment in march/20150317_CAR. 9 Aug 2017. sub-Saharan Africa. Tropical Med Int Health. 2014;19(8):968–77. Crellen et al. Conflict and Health (2019) 13:52 Page 11 of 11

47. Ssonko C, Gonzalez L, Mesic A, da Fonseca MS, Achar J, Safar N, et al. Delivering HIV care in challenging operating environments: the MSF experience towards differentiated models of care for settings with multiple basic health care needs. J Int AIDS Soc. 2017;20:21654. 48. Médecins Sans Frontières. Zemio, a new town under fire 2017. Available from: https://www.msf.org/central-african-republic-zemio-new-town-under-fire.1 Sept 2019.

Publisher’sNote Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.